[REQUEST] Input an image of the face we want to use as replace.

[ZeroCool22]

Instead, just using a prompt give us the option to use a face/picture we want to use as base to be replaced on the output pictures.

I saw a web some time ago hat does it, and was excellent.
So we can do faceswap not only for famous ppl.

[Echolink50]

I am looking for the same thing

[kex0]

I’m sorry but using pictures as a face replacement is out of scope of this project. This project is an extension for stable diffusion and is focused on generating faces with stable diffusion so if you want it to generate unique faces that it doesn't know, you should train it.

[Echolink50]

I’m sorry but using pictures as a face replacement is out of scope of this project. This project is an extension for stable diffusion and is focused on generating faces with stable diffusion so if you want it to generate unique faces that it doesn't know, you should train it.

I'm not sure what Zerocool22 is trying to do but I would just like a way to specify a folder of cropped faces to be used just like you can specify a folder of images you watch to have the face swapped on. Seems simple enough. Even if you can tell us the part of the code that tells the automatic1111 where to get the prompt generated face it could probably be modified to point to an existing directory instead of where ever the prompted face is stored before being swapped on to the image. Thanks

[kex0]

I just don't see how this would work. How would it select the right face from the folder that would fit the image. Assuming you even have the right picture. You would have to have a massive library of the face with all the different angles and lighting conditions otherwise it would not fit the rest of the image. At that point you can just train the SD model.

[Echolink50]

I just don't see how this would work. How would it select the right face from the folder that would fit the image. Assuming you even have the right picture. You would have to have a massive library of the face with all the different angles and lighting conditions otherwise it would not fit the rest of the image. At that point you can just train the SD model.

Same way the mask works. Just have the same name 0001.jpg face on 0001.jpg image

[kex0]

Oh gotcha. I'll think about it.

Still it would somehow have to do automatic color correction

[Echolink50]

In my use case the faces would already be cropped and color matched. I think trying to make the extension do all that is definitely asking to much. I only ask for this small feature because it seemed like it would be relatively easy to implement. Just a folder of faces the extension can look to if you don't want the text generated face.

[kex0]

Here try this branch:
https://github.com/kex0/batch-face-swap/tree/existing_face

A very rough piece of code that maybe does what you want. I have no way to test it.

[Echolink50]

Will try it out this evening. Thanks my friend

[Echolink50]

Here try this branch: https://github.com/kex0/batch-face-swap/tree/existing_face

A very rough piece of code that maybe does what you want. I have no way to test it.

The branch functions as expected. Additional settings for masks and things would be a welcomed addition. Thank you for your efforts

[2blackbar]

The existing face things is interesting idea, a code that detects closest face angle and tries to match them and paste onto image could be interesting or a fail cause light wouldnt match , but if its from same photshoot, who knows